Date report generated: 2024-05-01
The flow of study selection is shown in Figure 1. Studies included were published between 2014 and 2024. Overall, this analysis includes 14 studies containing 360 comparisons.
Figure 1
Table 1 below gives a summary of the included studies for the effect of exercise interventions. N represents an aggregate of animals contributing to outcomes reported from control and treatment groups, and if the same control group has contributed to more than one experiment, it will be counted twice.
| Study | Strain | Intervention | Outcome | N |
|---|---|---|---|---|
| AMIRI, 2021 | Wistar (rat) | exercise, 28 Days | Other behavioural | 112 |
| ~ | ~ | ~ | Other neurobiological | 28 |
| AMIRI, 2022 | Wistar (rat) | exercise, 28 Days | Fear memory | 20 |
| KOYUNCUOGLU, 2021 | Wistar (rat) | exercise, 42 Days | Freezing | 24 |
| ~ | ~ | ~ | Other behavioural | 120 |
| ~ | ~ | ~ | Other neurobiological | 240 |
| MIRJALILI, 2022 | Wistar (rat) | exercise, 28 Days | BDNF | 80 |
| ~ | ~ | ~ | Other behavioural | 80 |
| ~ | ~ | ~ | Other neurobiological | 160 |
| ~ | ~ | ~ | Stress response | 40 |
| MOHAMMADI, 2024 | Wistar (rat) | exercise, 28 Days | BDNF | 20 |
| ~ | ~ | ~ | Fear memory | 20 |
| ~ | ~ | ~ | Freezing | 20 |
| ~ | ~ | ~ | Locomotor | 60 |
| ~ | ~ | ~ | Other neurobiological | 100 |
| PATKI, 2014 | Wistar (rat) | exercise, 14 Days | Locomotor | 80 |
| ~ | ~ | ~ | Other behavioural | 80 |
| ~ | ~ | ~ | Stress response | 20 |
| SHAFIA, 2017 | Wistar (rat) | exercise, 28 Days | BDNF | 20 |
| ~ | ~ | ~ | Fear memory | 56 |
| ~ | ~ | ~ | Other behavioural | 98 |
| ~ | ~ | ~ | Other neurobiological | 40 |
| ~ | ~ | ~ | Stress response | 14 |
| SHAFIA, 2019 | Wistar (rat) | exercise, 3 Days | BDNF | 80 |
| ~ | ~ | ~ | Other behavioural | 40 |
| SHAFIA, 2022 | Wistar (rat) | exercise, 28 Days | Locomotor | 80 |
| ~ | ~ | ~ | Other neurobiological | 20 |
| ~ | ~ | ~ | Stress response | 20 |
| SHAFIA, 2023a | Wistar (rat) | exercise, 28 Days | BDNF | 28 |
| ~ | ~ | ~ | Fear memory | 14 |
| ~ | ~ | ~ | Other behavioural | 126 |
| SHAFIA, 2023b | Wistar (rat) | exercise, 28 Days | BDNF | 112 |
| ~ | ~ | ~ | Other behavioural | 112 |
| ~ | ~ | ~ | Stress response | 56 |
| SHAFIA, 2023c | Wistar (rat) | exercise, 28 Days | Other behavioural | 28 |
| ~ | ~ | ~ | Other neurobiological | 70 |
| YAKHKESHI, 2022 | Wistar (rat) | exercise, 28 Days | BDNF | 40 |
| ~ | ~ | ~ | Other behavioural | 180 |
| ~ | ~ | ~ | Other neurobiological | 80 |
| ~ | ~ | ~ | Stress response | 20 |
| ZHANG, 2020 | Sprague-dawley (rat) | exercise, 28 Days | BDNF | 20 |
| ~ | ~ | ~ | Freezing | 20 |
| ~ | ~ | ~ | Locomotor | 40 |
| ~ | ~ | ~ | Neurotransmitter levels | 200 |
| ~ | ~ | ~ | Other behavioural | 20 |
| ~ | ~ | ~ | Other neurobiological | 60 |
Table 2 below gives a summary of the included studies for the effect of model induction. N represents an aggregate of animals contributing to outcomes reported from control and treatment groups, and if the same control group has contributed to more than one experiment, those animals will be counted more than once.
| Study | Strain | Outcome | N |
|---|---|---|---|
| AMIRI, 2021 | Wistar (rat) | Other behavioural | 112 |
| AMIRI, 2022 | Wistar (rat) | Fear memory | 20 |
| KOYUNCUOGLU, 2021 | Wistar (rat) | Freezing | 12 |
| ~ | ~ | Other behavioural | 60 |
| ~ | ~ | Other neurobiological | 120 |
| MIRJALILI, 2022 | Wistar (rat) | BDNF | 80 |
| ~ | ~ | Other behavioural | 80 |
| ~ | ~ | Other neurobiological | 160 |
| ~ | ~ | Stress response | 40 |
| MOHAMMADI, 2024 | Wistar (rat) | BDNF | 20 |
| ~ | ~ | Fear memory | 20 |
| ~ | ~ | Freezing | 20 |
| ~ | ~ | Locomotor | 60 |
| ~ | ~ | Other neurobiological | 100 |
| PATKI, 2014 | Wistar (rat) | Locomotor | 80 |
| ~ | ~ | Other behavioural | 80 |
| ~ | ~ | Stress response | 20 |
| SHAFIA, 2017 | Wistar (rat) | BDNF | 20 |
| ~ | ~ | Fear memory | 56 |
| ~ | ~ | Other behavioural | 98 |
| ~ | ~ | Other neurobiological | 40 |
| ~ | ~ | Stress response | 28 |
| SHAFIA, 2019 | Wistar (rat) | BDNF | 80 |
| ~ | ~ | Other behavioural | 40 |
| SHAFIA, 2022 | Wistar (rat) | Locomotor | 80 |
| ~ | ~ | Other neurobiological | 20 |
| ~ | ~ | Stress response | 20 |
| SHAFIA, 2023a | Wistar (rat) | BDNF | 28 |
| ~ | ~ | Fear memory | 14 |
| ~ | ~ | Other behavioural | 70 |
| SHAFIA, 2023b | Wistar (rat) | BDNF | 56 |
| ~ | ~ | Other behavioural | 56 |
| ~ | ~ | Stress response | 28 |
| SHAFIA, 2023c | Wistar (rat) | Other behavioural | 28 |
| ~ | ~ | Other neurobiological | 70 |
| YAKHKESHI, 2022 | Wistar (rat) | BDNF | 40 |
| ~ | ~ | Other behavioural | 100 |
| ~ | ~ | Other neurobiological | 80 |
| ~ | ~ | Stress response | 20 |
| ZHANG, 2020 | Sprague-dawley (rat) | BDNF | 20 |
| ~ | ~ | Freezing | 20 |
| ~ | ~ | Locomotor | 40 |
| ~ | ~ | Neurotransmitter levels | 200 |
| ~ | ~ | Other behavioural | 20 |
| ~ | ~ | Other neurobiological | 60 |
References of included studies are located in the appendix. Included studies used 14 unique disease model induction procedures.
Within the literature we identified distinct categories of experiments and the data presented would allow several meta-analytical contrasts to be drawn:
Treatment vs control. These were experiments investigating the effect of performing exercise, reported in 170 experiments from 14 publications.
In these studies the:
Control group is a group of animals that is (1) subjected to a PTSD model induction paradigm and (2) administered a control treatment (vehicle) or no treatment.
Intervention group is a group of animals that is (1) subjected to a PTSD model induction paradigm and (2) performing exercise.
Sham group is a group of animals that is (1) not subjected to a PTSD model induction paradigm and (2) administered a control treatment (vehicle) or no treatment. These data are required to allow a ‘normalised mean difference’ effect size to be calculated, given by
\[ \frac{{\bar{\mu}_C - \bar{\mu}_T}}{{\bar{\mu}_C - \bar{\mu}_S}} \times 100 \]
where \(\bar{\mu}_C\), \(\bar{\mu}_T\), \(\bar{\mu}_S\) are the mean reported scores in the control, treatment, and sham groups respectively.
Effects of disease modelling. These are experiments investigating the effect of models of PTSD, reported in 143 experiments from 14 publications.
In these studies the:
Control group is a group of animals that is (1) not subjected to a PTSD model induction paradigm and (2) is administered a control treatment (vehicle) or no treatment.
Intervention group is a group of animals that is (1) subjected to a PTSD model induction paradigm and (2) is administered a control treatment (vehicle) or no treatment.
Outcomes with ≥2 independent effect sizes were considered for meta-analysis. In this iteration of the review, this includes other behavioural, other neurobiological, bdnf, stress response, fear memory, locomotor and freezing.
All analyses were conducted allowing for the following hierarchical levels in a random effects model, which accounts for features common to experimental contrasts such as a shared control group:
Level 1: Rodent strain - effect sizes measured across experiments using the same rodent strain.
Level 2: Study - effect sizes measured from different experiments presented in the same publication.
Level 3: Experiment - effect sizes measured in the same experiment within a study, where often a control group contributes to several effect sizes.
Each level for the hierarchy was only included in the model if more than 4 categories were present for at least one of these levels. Where more than 4 categories are not present for all levels, the variance attributable to levels with fewer than 5 categories is reported as zero.
The hierarchical grouping may therefore be considered thus: Strains of laboratory animals are included in several Studies, each of which can report one or more Experiments, and each Experiment is comprised of at least two Cohorts which are considered identical except for differing in the experimental manipulation (the Intervention) or not being exposed to the disease modelling procedures (a Sham cohort, these only being used to provide a baseline for outcome measures to allow Normalised Mean Difference meta-analysis). An Experiment can include several experimental contrasts, for instance where different doses of drugs are compared to the same control group. Several outcomes can be measured and reported from the same cohort of animals.
We constructed multilevel models without Hartung-Knapp adjustments as
these are not available for rma.mv class objects in the metafor package.
Instead, the model is set to test = "t" to use t- and
F-distributions for making inferences, and dfs="contain" to
improve the method of approximating degrees of freedom of these
distributions.
The scales and units used to measure outcomes in preclinical studies often differ between studies although they may measure the same underlying biological construct. The primary effect size used for meta-analysis of preclinical studies is therefore the standardised mean difference (SMD, Hedge’s g). For experiments testing the effects of interventions we also present a sensitivity analysis using normalised mean difference (NMD), where there are sufficient data for sham procedures to allow this. This analysis is not possible for studies of the effect of single prolonged stress (SPS).
14 studies (170 comparisons) investigated the effects of exercise versus control. The number of studies and individual effect sizes for each outcome were:
Locomotor outcomes: 4 studies and 13 comparisons in 2 strains
Fear Memory 4 studies and 7 comparisons in 1 strain
Freezing: 3 studies and 4 comparisons in 2 strains
Other behaviours: 11 studies and 50 comparisons in 2 strains
BDNF: 8 studies and 24 comparisons in 2 strains
Neurotransmitter levels: 1 studies and 10 comparisons in 1 strain
Biochemical stress responses: 6 studies and 10 comparisons in 1 strain
Other neurobiological outcomes: 9 studies and 52 comparisons in 2 strains
Multilevel analysis is only performed if there are 5 levels or more for at least one of Strain, Study and Experiment, and that is not the case here. 13 experimental comparisons were reported in 4 experiments from 4 publications and involving 2 different animal strain(s). We provide a conventional random effects model to illustrate the data. No subgroup analysis is performed. We also now include these locomotor outcomes under the ‘other behavioural’ heading below.
Multilevel analysis is only performed if there are 5 levels or more for at least one of Strain, Study and Experiment, and that is not the case here. 7 experimental comparisons were reported in 4 experiments from 4 publications and involving 1 different animal strain(s). We provide a conventional random effects model to illustrate the data. No subgroup analysis is performed. We also now include these fear memory outcomes under the ‘other behavioural’ heading below.
Multilevel analysis is only performed if there are 5 levels or more for at least one of Strain, Study and Experiment, and that is not the case here. 4 experimental comparisons were reported in 3 experiments from 3 publications and involving 2 different animal strain(s). We provide a conventional random effects model to illustrate the data. No subgroup analysis is performed. We also now include these Freezing outcomes under the ‘other behavioural’ heading below.
Figure 2.4.1 shows the risk of bias summary for studies investigating the effect of exercise on other behavioural outcomes in animals. The risk of bias assessment was performed using the SyRCLE RoB tool.
Figure 2.4.1
Figure 2.4.2 shows the reporting completeness summary for studies investigating the effect of exercise on other behavioural outcomes in animals. The reporting completeness assessment was performed using the ARRIVE guidelines.
Figure 2.4.2
The effect of exercise on other behavioural outcomes in animals using SMD as the effect size is shown in Figure 2.4.3. The pooled estimate for SMD across all individual comparisons is displayed as a diamond shape at the bottom of the plot. Dotted lines indicate the prediction interval of the pooled estimate.
Figure 2.4.3
Exercise interventions had a pooled effect on other behavioural outcomes induced by single prolonged stress of SMD = 1.333, (95% CI: 0.886 to 1.779; 95% PrI: -0.244 to 2.909).
74 experimental comparisons were reported in 19 experiments from 14 publications and involving 2 different animal strain(s).
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 14 | 0.184 |
| Study x Experiment | 19 | 0.306 |
For each outcome, the covariates of interest for subgroup analyses and meta-regressions were:
We also conducted subgroup analyses using (1) SyRCLE Risk of Bias and (2) ARRIVE reporting completeness assessment scores as covariates to evaluate their influence on effect size estimates. These were not specified in the study protocol, but evaluation of risk of bias is required for the Summary of Evidence table, and no studies were considered entirely at low risk of bias or of high reporting completeness to allow such a sensitivity analysis.
The significance (p-value) reported is that for a test of whether the moderators are significantly different one from another, rather than whether the effect is significantly different from 0.
Figure 2.4.4.1 displays the estimates for the pooled SMD’s when comparisons are stratified by sex of the animal. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by sex, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 2.4.4.1 - Effect of exercise on other behavioural outcomes by Sex
The p-value for the association between the sex of animal groups used and outcome reported was 0.121.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 14 | 0.41 |
| Study x Experiment | 19 | 0.145 |
Figure 2.4.4.2 displays the estimates for the pooled SMD when comparisons are stratified by whether the exercise was voluntary or forced (VoF). Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by exercise type, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 2.4.4.2 - Effect of exercise on other behavioural outcomes by voluntary or forced exercise
The p-value for the association between the VoF of animal groups used and outcome reported was 0.124.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 14 | 0.125 |
| Study x Experiment | 19 | 0.293 |
We provide a meta-regression of the number of weeks of treatment as a continuous variable.
Figure 2.4.4.3 - Effect of exercise on other behavioural outcomes by Duration of treatment
The p-value for the association between duration of treatment and outcome reported was 0.469.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0.017 |
| Study x Strain | 14 | 0.192 |
| Study x Strain x Experiment | 19 | 0.336 |
We provide a meta-regression where exercise intensity (expressed as m/min) is considered as a continuous variable.
Figure 2.4.4.4 - Effect of exercise on other behavioural outcomes by exercise intensity
The p-value for the association between exercise intensity and the outcome reported was 0.348.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0.22 |
| Study x Strain | 14 | 0.185 |
| Study x Strain x Experiment | 19 | 0.31 |
We provide a meta-regression where the total exercise as the product of the number of sessions (n), session duration (min), and session intensity (km/min), with total exercise expressed in km (n * min * km/min), is considered as a continuous variable.
Figure 2.4.4.5
The p-value for the association between total exercise and the outcome reported was 0.002.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0.076 |
| Study x Strain | 14 | 0.039 |
| Study x Strain x Experiment | 19 | 0.492 |
Figure 2.4.4.6 displays the estimates for the pooled SMD when comparisons are stratified by how many of the SyRCLE risk of bias assessment criteria (of which there are 10) the experiment met. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by SyRCLE Risk of Bias, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 2.4.4.6 - Effect of exercise on other behavioural outcomes by SyRCLE risk of bias criteria
The p-value for the association between SyRCLE Risks of Bias reporting and the outcome reported was 0.441.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 14 | 0.222 |
| Study x Experiment | 19 | 0.303 |
Figure 2.4.4.7 displays the estimates for the pooled SMD when comparisons are stratified by whether of not any of the SyRCLE Risk of bias domains were rated as low risk of bias. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by SyRCLE Risk of Bias, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 2.4.4.7 - Effect of exercise on other behavioural outcomes by low SyRCLE RoB
The p-value for the association between low SyRCLE Risks of Bias reporting and the outcome reported was 0.441.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 14 | 0.222 |
| Study x Experiment | 19 | 0.303 |
We provide a meta-regression where the number of ARRIVE items met is considered as a continuous variable.
Figure 2.4.4.8 - Effect of exercise on other behavioural outcomes by ARRIVE reporting completeness
The p-value for the association between ARRIVE reporting completeness and outcome reported was 0.519.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0 |
| Study x Strain | 14 | 0.234 |
| Study x Strain x Experiment | 19 | 0.309 |
The table below shows which of the covariates, if any, explain some of the heterogeneity observed in the effect of exercise on other behaviours. We present marginal R2, which measures the proportion of variance explained by including moderators in the model (the % change in the between-studies variance when the covariate is included in the model, in other words the % of the heterogeneity explained by the variable). The coefficients are derived form an RMA model fitted with an intercept (and so represent, for each category, the point estimate and 95% CIs of the effect in that category).
| Moderator | Category | \(\beta\) | 95% CI | Marginal R2 (%) |
|---|---|---|---|---|
| Overall effect | - | 1.333 | 0.886 to 1.779 | - |
| Sex | - | - | - | 9.5% |
| - | Female | 0.718 | -0.534 to 1.969 | - |
| - | Male | 1.325 | 0.143 to 2.507 | - |
| Voluntary or forced | - | - | - | 27.6% |
| - | Forced | 1.454 | 1.002 to 1.907 | - |
| - | Voluntary | 0.492 | -0.691 to 1.675 | - |
| Duration of treatment | - | - | - | 4.2% |
| - | per weeks of treatment increase | -0.129 | -0.503 to 0.246 | - |
| Exercise intensity | - | - | - | 1% |
| - | per unit (m/min) increase | -0.026 | -0.08 to 0.028 | - |
| Total exercise | - | - | - | 16% |
| - | per km increase | -0.06 | -0.097 to -0.022 | - |
| Risk of Bias | - | - | - | 5.4% |
| - | 0 criteria met | 1.269 | 0.775 to 1.764 | - |
| - | 1 criteria met | 1.795 | 0.444 to 3.147 | - |
| Reporting completeness | - | - | - | 3.7% |
| - | per unit increase | -0.089 | -0.381 to 0.203 | - |
We examine the robustness of the findings for other behaviours by performing the following sensitivity analyses.
In the previous analyses for the effect of exercise on other behavioural outcomes, we imputed a \(\rho\) value - the imputed within-study correlation between observed effect sizes - of 0.5. Here, we examine the effect of imputing \(\rho\) values of 0.2 and 0.8.
When the \(\rho\) value is assumed to be 0.2, exercise had a pooled effect on other behavioural outcomes of SMD = 1.59 (95% CI: 1.08 to 2.1; 95% PrI: -0.29 to 3.47).
When the \(\rho\) value is assumed to be 0.8, exercise had a pooled effect on other behavioural outcomes of SMD = 0.65 (95% CI: 0.04 to 1.27; 95% PrI: -1.71 to 3.02).
For reference the pooled effect size when rho is assumed to be 0.5 is 1.33 (95% CI: 0.89 to 1.78).
For behavioural outcomes, an NMD was calculable for 74 out of 74 comparisons, i.e. 100 % of comparisons.
The effect of exercise on other behaviours in animals using NMD as the effect size is shown in Figure 2.4.5. The pooled estimate for NMD across all individual comparisons is displayed as a diamond shape at the bottom of the plot. Dotted lines indicate the prediction interval of the pooled estimate.
Figure 2.4.5
Exercise had a pooled effect on other behaviours of NMD = 97.38 (95% CI: -190.65 to 385.42) with a prediction interval of -1115.57 to 1310.33). For reference the pooled effect size for SMD was 1.33 (95% CI: 0.89 to 1.78).
74 experimental comparisons were reported in 19 experiments reported from 14 publications and involving 2 different animal strains.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 6.37 |
| Study x Strain | 14 | 2222.83 |
| Study x Strain x Experiment | 19 | 6369.77 |
Because of the relationship between SMD effect sizes and variance inherent in their calculation, where study size is small the standard approach to seeking evidence of small-study effects (regression based tests including Egger’s regression test for multilevel meta-analysis) can lead to over-estimation of small-study effect (see for instance 10.7554/eLife.24260). To address this we used Egger’s regression test for multilevel meta-analysis, with regression of SMD effect size against 1/√N, where N is the total number of animals involved in an experiment.
Egger regression based on 74 studies of modelling of depression where sucrose preference was measured showed a coefficient for a small-study effect of -4.18 (95% CI: -23.8 to 15.43; p = 0.650).
Figure 2.5.1 shows the risk of bias summary for studies investigating the effect of exercise on BDNF in animals. The risk of bias assessment was performed using the SyRCLE RoB tool.
Figure 2.5.1
Figure 2.5.2 shows the reporting completeness summary for studies investigating the effect of exercise on BDNF in animals. The reporting completeness assessment was performed using the ARRIVE guidelines.
Figure 2.5.2
The effect of exercise on BDNF in animals using SMD as the effect size is shown in Figure 2.5.3. The pooled estimate for SMD across all individual comparisons is displayed as a diamond shape at the bottom of the plot. Dotted lines indicate the prediction interval of the pooled estimate.
Figure 2.5.3
Exercise had a pooled effect on BDNF of SMD = 1.79 , (95% CI: 0.56 to 3.01; 95% PrI: -1.63 to 5.21).
24 experimental comparisons were reported in 11 experiments from 8 publications and involving 2 different animal strain(s).
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 8 | 1.82 |
| Study x Experiment | 11 | 0 |
For each outcome, the covariates of interest for subgroup analyses and meta-regressions were:
We also conducted subgroup analyses using (1) SyRCLE Risk of Bias and (2) ARRIVE reporting completeness assessment scores as covariates to evaluate their influence on effect size estimates. These were not specified in the study protocol, but evaluation of risk of bias is required for the Summary of Evidence table, and no studies were considered entirely at low risk of bias or of high reporting completeness to allow such a sensitivity analysis.
The significance (p-value) reported is that for a test of whether the moderators are significantly different one from another, rather than whether the effect is significantly different from 0.
Figure 2.5.4.1 displays the estimates for the pooled SMD when comparisons are stratified by sex of the animal. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by sex, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 2.5.4.1 - Effect of exercise on BDNF by Sex
The p-value for the association between the sex of animal groups used and outcome reported was 0.847.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 8 | 1.892 |
| Study x Experiment | 11 | 0 |
Figure 2.5.4.2 displays the estimates for the pooled SMD when comparisons are stratified by whether the exercise was voluntary or forced (VoF). Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by exercise type, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 2.5.4.2 - Effect of exercise on other BDNF by voluntary or forced exercise
The p-value for the association between the VoF of animal groups used and outcome reported was 0.868.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 8 | 2.228 |
| Study x Experiment | 11 | 0 |
We provide a meta-regression of the number of weeks of treatment as a continuous variable.
Figure 2.5.4.3 - Effect of exercise on BDNF by duration of treatment
The p-value for the association between duration of treatment and outcome reported was 0.428.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0 |
| Study x Strain | 8 | 1.992 |
| Study x Strain x Experiment | 11 | 0 |
We provide a meta-regression where exercise intensity (expressed at m/min) is considered as a continuous variable.
Figure 2.5.4.4 - Effect of exercise on BDNF by exercise intensity
The p-value for the association between exercise intensity and outcome reported was 0.007.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0 |
| Study x Strain | 8 | 2.811 |
| Study x Strain x Experiment | 11 | 0 |
We provide a meta-regression where the total exercise as the product of the number of sessions, session duration, and session intensity, expressed in km, is considered as a continuous variable.
Figure 2.5.4.5 - Effect of exercise on BDNF by total exercise
The p-value for the association between total exercise and outcome reported was 0.006.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0 |
| Study x Strain | 8 | 2.711 |
| Study x Strain x Experiment | 11 | 0 |
No studies met any RoB criteria.
No studies met any RoB criteria.
We provide a meta-regression where the number of ARRIVE items met is considered as a continuous variable.
Figure 2.5.4.6 - Effect of exercise on BDNF by ARRIVE reporting completeness
The p-value for the association between ARRIVE reporting completeness and outcome reported was 0.756.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0 |
| Study x Strain | 8 | 2.223 |
| Study x Strain x Experiment | 11 | 0 |
The table below shows which of the covariates, if any, explain some of the heterogeneity observed in the effect of exercise on BDNF. We present marginal R2, which measures the proportion of variance explained by including moderators in the model (the % change in the between-studies variance when the covariate is included in the model, in other words the % of the heterogeneity explained by the variable). The coefficients are derived form an RMA model fitted with an intercept (and so represent, for each category, the point estimate and 95% CIs of the effect in that category).
| Moderator | Category | \(\beta\) | 95% CI | Marginal R2 (%) |
|---|---|---|---|---|
| Overall effect | - | 1.789 | 0.564 to 3.013 | - |
| Sex | - | - | - | 0.1% |
| - | Female | 1.751 | 0.469 to 3.033 | - |
| - | Male | 1.832 | 0.559 to 3.106 | - |
| Voluntary or forced | - | - | - | 0.3% |
| - | Forced | 1.885 | -19.253 to 23.024 | - |
| - | Voluntary | 1.305 | -26.738 to 29.349 | - |
| Duration of treatment | - | - | - | 11.2% |
| - | per weeks of treatment increase | 0.369 | -0.695 to 1.433 | - |
| Exercise intensity | - | - | - | 8.7% |
| - | per unit (m/min) increase | -0.18 | -0.304 to -0.055 | - |
| Total exercise | - | - | - | 7.8% |
| - | per km increase | -0.077 | -0.129 to -0.024 | - |
| Reporting completeness | - | - | - | 1.4% |
| - | per unit increase | 0.117 | -0.762 to 0.996 | - |
We examine the robustness of the findings for BDNF by performing the following sensitivity analyses.
In the previous analyses for the effect of exercise on BDNF, we imputed a \(\rho\) value - the imputed within-study correlation between observed effect sizes - of 0.5. Here, we examine the effect of imputing \(\rho\) values of 0.2 and 0.8.
When the \(\rho\) value is assumed to be 0.2, exercise had a pooled effect on BDNF of SMD = 2.13 (95% CI: 0.64 to 3.62; 95% PrI: -2.15 to 6.4).
When the \(\rho\) value is assumed to be 0.8, exercise had a pooled effect on BDNF of SMD = 1.08 (95% CI: 0.04 to 2.12; 95% PrI: -1.81 to 3.97).
For reference the pooled effect size when rho is assumed to be 0.5 is 1.79 (95% CI: 0.56 to 3.01).
For behavioural outcomes, an NMD was calculable for 24 out of 24 comparisons, i.e. 100 % of comparisons.
The effect of exercise on BDNF outcomes in animals using NMD as the effect size is shown in Figure 2.5.5. The pooled estimate for NMD across all individual comparisons is displayed as a diamond shape at the bottom of the plot. Dotted lines indicate the prediction interval of the pooled estimate.
Figure 2.5.5
Exercise had a pooled effect on BDNF outcomes of NMD = 88.67 (95% CI: -169.96 to 347.3) with a prediction interval of -688.58 to 865.92). For reference the pooled effect size for SMD was 1.79 (95% CI: 0.56 to 3.01).
24 experimental comparisons were reported in 11 experiments reported from 8 publications and involving 2 different animal strains.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 1.67 |
| Study x Strain | 8 | 3111.68 |
| Study x Strain x Experiment | 11 | 214.19 |
Because of the relationship between SMD effect sizes and variance inherent in their calculation, where study size is small the standard approach to seeking evidence of small-study effects (regression based tests including Egger’s regression test for multilevel meta-analysis) can lead to over-estimation of small-study effect (see for instance 10.7554/eLife.24260). To address this we used Egger’s regression test for multilevel meta-analysis, with regression of SMD effect size against 1/√N, where N is the total number of animals involved in an experiment.
Egger regression based on 24 studies of modelling of depression where sucrose preference was measured showed a coefficient for a small-study effect of -5.94 (95% CI: -48.46 to 36.59; p = 0.744).
Figure 2.6.1 shows the risk of bias summary for studies investigating the effect of exercise on biological stress response in animals. The risk of bias assessment was performed using the SyRCLE RoB tool.
Figure 2.6.1
Figure 2.6.2 shows the reporting completeness summary for studies investigating the effect of exercise on biological stress response in animals. The reporting completeness assessment was performed using the ARRIVE guidelines.
Figure 2.6.2
The effect of exercise on biological stress response in animals using SMD as the effect size is shown in Figure 2.6.3. The pooled estimate for SMD across all individual comparisons is displayed as a diamond shape at the bottom of the plot. Dotted lines indicate the prediction interval of the pooled estimate.
Figure 2.6.3
Exercise interventions had a pooled effect on biological stress response induced by single prolonged stress of SMD = 2.03, (95% CI: -1.78 to 5.84; 95% PrI: -7.88 to 11.93).
10 experimental comparisons were reported in 8 experiments from 6 publications and involving 1 different animal strain(s).
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 6 | 12.65 |
| Study x Experiment | 8 | 0 |
For each outcome, the covariates of interest for subgroup analyses and meta-regressions were:
We also conducted subgroup analyses using (1) SyRCLE Risk of Bias and (2) ARRIVE reporting completeness assessment scores as covariates to evaluate their influence on effect size estimates. These were not specified in the study protocol, but evaluation of risk of bias is required for the Summary of Evidence table, and no studies were considered entirely at low risk of bias or of high reporting completeness to allow such a sensitivity analysis.
The significance (p-value) reported is that for a test of whether the moderators are significantly different one from another, rather than whether the effect is significantly different from 0.
Figure 2.6.4.1 displays the estimates for the pooled SMD when comparisons are stratified by sex of the animal. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by sex, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 2.6.4.1 - Effect of exercise on biological stress response by Sex
The p-value for the association between the sex of animal groups used and biological stress response was 0.631.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 6 | 12.575 |
| Study x Experiment | 8 | 0 |
All studies used forced exercise.
We provide a meta-regression of the number of weeks of treatment as a continuous variable.
Figure 2.6.4.2 - Effect of exercise on biological stress response by duration of treatment
The p-value for the association between duration of treatment and biological stress response was 0.878.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 1 | 0 |
| Study x Strain | 6 | 16.169 |
| Study x Strain x Experiment | 8 | 0 |
We provide a meta-regression where exercise intensity is considered as a continuous variable.
Figure 2.6.4.3 - Effect of exercise on biological stress response by exercise intensity
The p-value for the association between exercise intensity and biological stress response was 0.005.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 1 | 0 |
| Study x Strain | 6 | 12.248 |
| Study x Strain x Experiment | 8 | 0 |
We provide a meta-regression where the total exercise as the product of the number of sessions, session duration, and session intensity, expressed in km, is considered as a continuous variable.
Figure 2.6.4.4 - Effect of exercise on biological stress response by total exercise
The p-value for the association between total exercise and biological stress response was 0.006.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 1 | 0 |
| Study x Strain | 6 | 12.251 |
| Study x Strain x Experiment | 8 | 0 |
Figure 2.6.4.5 displays the estimates for the pooled SMD when comparisons are stratified by how many of the SyRCLE risk of bias assessment criteria (of which there are 10) that the experiment met. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by SyRCLE Risk of Bias, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 2.6.4.5 - Effect of exercise on biological stress response by SyRCLE RoB criteria
The p-value for the association between SyRCLE Risks of Bias reporting and biological stress response was 0.878.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 6 | 16.169 |
| Study x Experiment | 8 | 0 |
Figure 2.6.4.6 displays the estimates for the pooled SMD when comparisons are stratified by whether of not any of the SyRCLE Risk of bias domains were rated as low risk of bias. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by SyRCLE Risk of Bias, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 2.6.4.6 - Effect of exercise on biological stress response by low SyRCLE RoB
The p-value for the association between low SyRCLE Risks of Bias reporting and biological stress response was 0.878.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 6 | 16.169 |
| Study x Experiment | 8 | 0 |
We provide a meta-regression where the number of ARRIVE items met is considered as a continuous variable.
Figure 2.6.4.7 - Effect of exercise on biological stress response by ARRIVE reporting completeness
The p-value for the association between ARRIVE reporting completeness and biological stress response was 0.704.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 1 | 0 |
| Study x Strain | 6 | 15.625 |
| Study x Strain x Experiment | 8 | 0 |
The table below shows which of the covariates, if any, explain some of the heterogeneity observed in the effect of exercise on biological stress response. We present marginal R2, which measures the proportion of variance explained by including moderators in the model (the % change in the between-studies variance when the covariate is included in the model, in other words the % of the heterogeneity explained by the variable). The coefficients are derived form an RMA model fitted with an intercept (and so represent, for each category, the point estimate and 95% CIs of the effect in that category).
| Moderator | Category | \(\beta\) | 95% CI | Marginal R2 (%) |
|---|---|---|---|---|
| Overall effect | - | 2.028 | -1.783 to 5.839 | - |
| Sex | - | - | - | 0.3% |
| - | Female | 2.23 | -1.518 to 5.978 | - |
| - | Male | 1.88 | -1.807 to 5.568 | - |
| Duration of treatment | - | - | - | 0.3% |
| - | per weeks of treatment increase | 0.364 | -5.81 to 6.537 | - |
| Exercise intensity | - | - | - | 4.4% |
| - | per unit (m/min) increase | -0.315 | -0.508 to -0.122 | - |
| Total exercise | - | - | - | 3% |
| - | per km increase | -0.132 | -0.212 to -0.051 | - |
| Risk of Bias | - | - | - | 0.3% |
| - | 0 criteria met | 2.166 | -2.92 to 7.252 | - |
| - | 1 criteria met | 1.439 | -9.812 to 12.69 | - |
| Reporting completeness | - | - | - | 2.8% |
| - | per unit increase | -0.395 | -3.081 to 2.29 | - |
We examine the robustness of the findings for biological stress response by performing the following sensitivity analyses.
In the previous analyses for the effect of exercise on biological stress response, we imputed a \(\rho\) value - the imputed within-study correlation between observed effect sizes - of 0.5. Here, we examine the effect of imputing \(\rho\) values of 0.2 and 0.8.
When the \(\rho\) value is assumed to be 0.2, exercise had a pooled effect on biological stress response of SMD = 2.04 (95% CI: -1.76 to 5.85; 95% PrI: -7.84 to 11.93).
When the \(\rho\) value is assumed to be 0.8, exercise had a pooled effect on biological stress response of SMD = 1.96 (95% CI: -1.88 to 5.81; 95% PrI: -8.04 to 11.96).
For reference the pooled effect size when rho is assumed to be 0.5 is 2.03 (95% CI: -1.78 to 5.84).
For stress response outcomes, an NMD was calculable for 10 out of 10 comparisons, i.e. 100 % of comparisons.
The effect of exercise on stress response outcomes in animals using NMD as the effect size is shown in Figure 2.6.5. The pooled estimate for NMD across all individual comparisons is displayed as a diamond shape at the bottom of the plot. Dotted lines indicate the prediction interval of the pooled estimate.
Figure 2.6.5
Exercise had a pooled effect on stress response outcomes of NMD = 71.47 (95% CI: -85.25 to 228.2) with a prediction interval of -341.69 to 484.64). For reference the pooled effect size for SMD was 2.03 (95% CI: -1.78 to 5.84).
10 experimental comparisons were reported in 8 experiments reported from 6 publications and involving 1 different animal strains.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 1 | 0 |
| Study x Strain | 6 | 883.02 |
| Study x Strain x Experiment | 8 | 22.19 |
Because of the relationship between SMD effect sizes and variance inherent in their calculation, where study size is small the standard approach to seeking evidence of small-study effects (regression based tests including Egger’s regression test for multilevel meta-analysis) can lead to over-estimation of small-study effect (see for instance 10.7554/eLife.24260). To address this we used Egger’s regression test for multilevel meta-analysis, with regression of SMD effect size against 1/√N, where N is the total number of animals involved in an experiment.
Egger regression based on 10 studies of modelling of depression where other neurobiological outcomes was measured showed a coefficient for a small-study effect of -111.34 (95% CI: -271.68 to 48.99; p = 0.126).
Multilevel analysis is only performed if there are 5 levels or more for at least one of Strain, Study and Experiment, and that is not the case here. 10 experimental comparisons were reported in 1 experiments from 1 publications and involving 1 different animal strain(s). We provide a conventional random effects model to illustrate the data. No subgroup analysis is performed. We also now include these Freezing outcomes under the ‘other neurobiological’ heading below.
Figure 2.8.1 shows the risk of bias summary for studies investigating the effect of exercise on other neurobiological outcomes in animals. The risk of bias assessment was performed using the SyRCLE RoB tool.
Figure 2.8.1
Figure 2.8.2 shows the reporting completeness summary for studies investigating the effect of exercise on other neurobiological outcomes in animals. The reporting completeness assessment was performed using the ARRIVE guidelines.
Figure 2.8.2
The effect of exercise on other neurobiological outcomes in animals using SMD as the effect size is shown in Figure 2.8.3. The pooled estimate for SMD across all individual comparisons is displayed as a diamond shape at the bottom of the plot. Dotted lines indicate the prediction interval of the pooled estimate.
Figure 2.8.3
Exercise interventions had a pooled effect on other neurobiological outcomes induced by single prolonged stress of SMD = 1.1, (95% CI: -0.18 to 2.38; 95% PrI: -2.79 to 4.99).
62 experimental comparisons were reported in 11 experiments from 9 publications and involving 2 different animal strain(s).
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 9 | 2.53 |
| Study x Experiment | 11 | 0 |
For each outcome, the covariates of interest for subgroup analyses and meta-regressions were:
We also conducted subgroup analyses using (1) SyRCLE Risk of Bias and (2) ARRIVE reporting completeness assessment scores as covariates to evaluate their influence on effect size estimates. These were not specified in the study protocol, but evaluation of risk of bias is required for the Summary of Evidence table, and no studies were considered entirely at low risk of bias or of high reporting completeness to allow such a sensitivity analysis.
The significance (p-value) reported is that for a test of whether the moderators are significantly different one from another, rather than whether the effect is significantly different from 0.
Figure 2.8.4.1 displays the estimates for the pooled SMD when comparisons are stratified by sex of the animal. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by sex, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 2.8.4.1 - Effect of exercise on other neurobiological outcomes by Sex
The p-value for the association between the sex of animal groups used and other neurobiological outcomes was 0.935.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 9 | 2.677 |
| Study x Experiment | 11 | 0 |
Figure 2.8.4.2 displays the estimates for the pooled SMD when comparisons are stratified by whether the exercise was voluntary or forced. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by sex, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 2.8.4.2 - Effect of exercise on other neurobiological outcomes by voluntary or forced exercise
The p-value for the association between the VoF of animal groups used and other neurobiological outcomes was 0.181.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 9 | 2.146 |
| Study x Experiment | 11 | 0 |
We provide a meta-regression of the number of weeks of treatment as a continuous variable.
Figure 2.8.4.3 - Effect of exercise on other neurobiological outcomes by duration of treatment
The p-value for the association between duration of treatment and other neurobiological outcomes was 0.291.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0 |
| Study x Strain | 9 | 2.444 |
| Study x Strain x Experiment | 11 | 0 |
We provide a meta-regression where exercise intensity is considered as a continuous variable.
Figure 2.8.4.4 - Effect of exercise on other neurobiological outcomes by exercise intensity
The p-value for the association between exercise intensity and outcome reported was 0.04.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0 |
| Study x Strain | 9 | 2.56 |
| Study x Strain x Experiment | 11 | 0 |
We provide a meta-regression where the total exercise as the product of the number of sessions, session duration, and session intensity, expressed in km, is considered as a continuous variable.
Figure 2.8.4.5 - Effect of exercise on other neurobiological outcomes by total exercise
The p-value for the association between total exercise and other neurobiological outcomes was 0.063.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0 |
| Study x Strain | 9 | 2.906 |
| Study x Strain x Experiment | 11 | 0 |
No studies met any RoB criteria.
No studies met any RoB criteria.
We provide a meta-regression where the number of ARRIVE items met is considered as a continuous variable.
Figure 2.8.4.6 - Effect of exercise on other neurobiological outcomes by ARRIVE reporting completeness
The p-value for the association between ARRIVE reporting completeness and other neurobiological outcomes was 0.259.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0 |
| Study x Strain | 9 | 2.482 |
| Study x Strain x Experiment | 11 | 0 |
The table below shows which of the covariates, if any, explain some of the heterogeneity observed in the effect of exercise on other neurobiological outcomes. We present marginal R2, which measures the proportion of variance explained by including moderators in the model (the % change in the between-studies variance when the covariate is included in the model, in other words the % of the heterogeneity explained by the variable). The coefficients are derived form an RMA model fitted with an intercept (and so represent, for each category, the point estimate and 95% CIs of the effect in that category).
| Moderator | Category | \(\beta\) | 95% CI | Marginal R2 (%) |
|---|---|---|---|---|
| Overall effect | - | 1.101 | -0.181 to 2.384 | - |
| Sex | - | - | - | 0% |
| - | Female | 1.133 | -0.375 to 2.642 | - |
| - | Male | 1.074 | -0.451 to 2.599 | - |
| Voluntary or forced | - | - | - | 28% |
| - | Forced | 1.513 | 0.12 to 2.906 | - |
| - | Voluntary | -0.305 | -2.839 to 2.228 | - |
| Duration of treatment | - | - | - | 26% |
| - | per weeks of treatment increase | -0.983 | -3.021 to 1.054 | - |
| Exercise intensity | - | - | - | 2.2% |
| - | per unit (m/min) increase | 0.055 | 0.002 to 0.108 | - |
| Total exercise | - | - | - | 2.7% |
| - | per km increase | 0.054 | -0.003 to 0.112 | - |
| Reporting completeness | - | - | - | 12.8% |
| - | per unit increase | -0.388 | -1.134 to 0.359 | - |
We examine the robustness of the findings for the primary outcome by performing the following sensitivity analyses.
In the previous analyses for the effect of exercise on other neurobiological outcomes, we imputed a \(\rho\) value - the imputed within-study correlation between observed effect sizes - of 0.5. Here, we examine the effect of imputing \(\rho\) values of 0.2 and 0.8.
When the \(\rho\) value is assumed to be 0.2, exercise had a pooled effect on other neurobiological outcomes of SMD = 1.69 (95% CI: 0.62 to 2.76; 95% PrI: -1.53 to 4.91).
When the \(\rho\) value is assumed to be 0.8, exercise had a pooled effect on other neurobiological outcomes of SMD = 0.09 (95% CI: -1.73 to 1.91; 95% PrI: -5.54 to 5.72).
For reference the pooled effect size when rho is assumed to be 0.5 is 1.1 (95% CI: -0.18 to 2.38.
For other neurobiological outcomes, an NMD was calculable for 60 out of 62 comparisons, i.e. 96.77 % of comparisons.
The effect of exercise on other neurobiological outcomes in animals using NMD as the effect size is shown in Figure 2.8.5. The pooled estimate for NMD across all individual comparisons is displayed as a diamond shape at the bottom of the plot. Dotted lines indicate the prediction interval of the pooled estimate.
Figure 2.8.5
Exercise had a pooled effect on stress response outcomes of NMD = 73.42 (95% CI: -102.47 to 249.3) with a prediction interval of -452.97 to 599.81). For reference the pooled effect size for SMD was 1.1 (95% CI: -0.18 to 2.38).
60 experimental comparisons were reported in 9 experiments reported from 8 publications and involving 2 different animal strains.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0.63 |
| Study x Strain | 8 | 1484.7 |
| Study x Strain x Experiment | 9 | 39.34 |
Because of the relationship between SMD effect sizes and variance inherent in their calculation, where study size is small the standard approach to seeking evidence of small-study effects (regression based tests including Egger’s regression test for multilevel meta-analysis) can lead to over-estimation of small-study effect (see for instance 10.7554/eLife.24260). To address this we used Egger’s regression test for multilevel meta-analysis, with regression of SMD effect size against 1/√N, where N is the total number of animals involved in an experiment.
Egger regression based on 62 studies of modelling of depression where other neurobiological outcomes was measured showed a coefficient for a small-study effect of -27.2 (95% CI: -62.64 to 8.24; p = 0.112).
14 studies (143 comparisons) investigated the effects of model induction. The number of studies and individual effect sizes for each outcome were:
Locomotor outcomes: 4 studies and 13 comparisons in 2 strains
Fear Memory 4 studies and 7 comparisons in 1 strain
Freezing: 3 studies and 3 comparisons in 2 strains
Other behaviours: 11 studies and 41 comparisons in 2 strains
BDNF: 8 studies and 20 comparisons in 2 strains
Neurotransmitter levels: 1 studies and 10 comparisons in 1 strain
Biochemical stress responses: 6 studies and 9 comparisons in 1 strain
Other neurobiological outcomes: 8 studies and 40 comparisons in 2 strains
Multilevel analysis is only performed if there are 5 levels or more for at least one of Strain, Study and Experiment, and that is not the case here. 13 experimental comparisons were reported in 4 experiments from 4 publications and involving 2 different animal strain(s). We provide a conventional random effects model to illustrate the data. No subgroup analysis is performed. We also now include these locomotor outcomes under the ‘other behavioural’ heading below.
Multilevel analysis is only performed if there are 5 levels or more for at least one of Strain, Study and Experiment, and that is not the case here. 7 experimental comparisons were reported in 4 experiments from 4 publications and involving 1 different animal strain(s). We provide a conventional random effects model to illustrate the data. No subgroup analysis is performed. We also now include these fear memory outcomes under the ‘other behavioural’ heading below.
Multilevel analysis is only performed if there are 5 levels or more for at least one of Strain, Study and Experiment, and that is not the case here. 3 experimental comparisons were reported in 3 experiments from 3 publications and involving 2 different animal strain(s). We provide a conventional random effects model to illustrate the data. No subgroup analysis is performed. We also now include these freezing outcomes under the ‘other behavioural’ heading below.
Figure 3.4.1 shows the risk of bias summary for studies investigating the effect of single prolonged stress on other behavioural outcomes in animals. The risk of bias assessment was performed using the SyRCLE RoB tool.
Figure 3.4.1
Figure 3.4.2 shows the reporting completeness summary for studies investigating the effect of single prolonged stress on sucrose preference in animals. The reporting completeness assessment was performed using the ARRIVE guidelines.
Figure 3.4.2
The effect of single prolonged stress on other behaviours in animals using SMD as the effect size is shown in Figure 3.4.3. The pooled estimate for SMD across all individual comparisons is displayed as a diamond shape at the bottom of the plot. Dotted lines indicate the prediction interval of the pooled estimate.
Figure 3.4.3
Single prolonged stress had a pooled effect on other behaviours of SMD = -1.68, (95% CI: -2.52 to -0.85; 95% PrI: -4.74 to 1.37).
64 experimental comparisons were reported in 19 experiments from 14 publications and involving 2 different animal strain(s).
The following table structure is used throughout this report and is used to show the different levels contributing to that analysis, the number of unique categories in those levels, and the variance contributed by that level of analysis. Because levels are only included in the analysis where there are five or more unique categories, for some analyses the number of categories is 0, and the variance attributed to those levels in not applicable. Because the model is hierarchical, where for instance there are Studies which include different Strains, the number of categories for Study x Strain will exceed the number of Studies (by which we mean unique publications) referred to in the text.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 14 | 1.76 |
| Study x Experiment | 19 | 0.08 |
For each outcome, the covariates of interest for subgroup analyses and meta-regressions were:
We also conducted subgroup analyses using (1) SyRCLE Risk of Bias and (2) ARRIVE reporting completeness assessment scores as covariates to evaluate their influence on effect size estimates. These were not specified in the study protocol, but evaluation of risk of bias is required for the Summary of Evidence table, and no studies were considered entirely at low risk of bias or of high reporting completeness to allow such a sensitivity analysis
The significance (p-value) reported is that for a test of whether the moderators are significantly different one from another, rather than whether the effect is significantly different from 0.
Figure 3.4.4.1 displays the estimates for the pooled SMD’s when comparisons are stratified by sex of the animal. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by sex, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 3.4.4.1 - Effect of single prolonged stress on other behavioural outcomes by Sex
The p-value for the association between the sex of animal groups used and outcome reported was 0.211.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 14 | 2.079 |
| Study x Experiment | 19 | 0 |
Figure 3.4.4.3 displays the estimates for the pooled SMD’s when comparisons are stratified by how many of the SyRCLE risk of bias assessment criteria (of which there are 10) that the experiment met. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by SyRCLE Risk of Bias, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 3.4.4.3 - Effect of single prolonged stress on other behavioural outcomes by SyRCLE RoB criteria
The p-value for the association between SyRCLE Risks of Bias reporting and outcome reported was 0.143.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 14 | 1.941 |
| Study x Experiment | 19 | 0.083 |
Figure 3.4.4.4 displays the estimates for the pooled SMD’s when comparisons are stratified by whether of not any of the SyRCLE Risk of bias domains were rated as low risk of bias. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by SyRCLE Risk of Bias, is displayed as a diamond shape at the bottom of the plot.
Figure 3.4.4.4 - Effect of single prolonged stress on other behavioural outcomes by low SyRCLE RoB
The p-value for the association between low SyRCLE Risks of Bias reporting and outcome reported was 0.143.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 14 | 1.941 |
| Study x Experiment | 19 | 0.083 |
We provide a meta-regression where the number of ARRIVE items met is considered as a continuous variable.
Figure 3.4.4.5 - Effect of single prolonged stress on other behavioural outcomes by ARRIVE reporting completeness
The p-value for the association between ARRIVE reporting completeness and outcome reported was 0.33.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0 |
| Study x Strain | 14 | 2.01 |
| Study x Strain x Experiment | 19 | 0.082 |
The table below shows which of the covariates, if any, explain some of the heterogeneity observed in the effect of model induction on other behavioural outcomes. We present marginal R2, which measures the proportion of variance explained by including moderators in the model (the % change in the between-studies variance when the covariate is included in the model, in other words the % of the heterogeneity explained by the variable). The coefficients are derived form an RMA model fitted with an intercept (and so represent, for each category, the point estimate and 95% CIs of the effect in that category).
| Moderator | Category | \(\beta\) | 95% CI | Marginal R2 (%) |
|---|---|---|---|---|
| Overall effect | - | -1.684 | -2.52 to -0.848 | - |
| Sex | - | - | - | 2.1% |
| - | Female | -1.311 | -2.667 to 0.046 | - |
| - | Male | -1.768 | -3.083 to -0.454 | - |
| Risk of Bias | - | - | - | 18.9% |
| - | 0 criteria met | -1.467 | -2.403 to -0.532 | - |
| - | 1 criteria met | -3.43 | -5.995 to -0.865 | - |
| Reporting completeness | - | - | - | 6.6% |
| - | per unit increase | 0.253 | -0.289 to 0.794 | - |
We examine the robustness of the findings for other behaviours by performing the following sensitivity analyses.
In the previous analyses for the effect of single prolonged stress on other behaviours, we imputed a \(\rho\) value - the imputed within-study correlation between observed effect sizes - of 0.5. Here, we examine the effect of imputing \(\rho\) values of 0.2 and 0.8.
When the \(\rho\) value is assumed to be 0.2, single prolonged stress had a pooled effect on other behaviours of SMD = -1.92 (95% CI: -2.64 to -1.2; 95% PrI: -4.55 to 0.71).
When the \(\rho\) value is assumed to be 0.8, single prolonged stress had a pooled effect on other behaviours of SMD = -1.11 (95% CI: -2.28 to 0.06; 95% PrI: -5.54 to 3.33).
For reference the pooled effect size when rho is assumed to be 0.5 is -1.68 (95% CI: -2.52 to -0.85).
Because of the relationship between SMD effect sizes and variance inherent in their calculation, where study size is small the standard approach to seeking evidence of small-study effects (regression based tests including Egger’s regression test for multilevel meta-analysis) can lead to over-estimation of small-study effect (see for instance 10.7554/eLife.24260). To address this we used Egger’s regression test for multilevel meta-analysis, with regression of SMD effect size against 1/√N, where N is the total number of animals involved in an experiment.
Egger regression based on 64 studies of modelling of depression where sucrose preference was measured showed a coefficient for a small-study effect of -4.8 (95% CI: -42.22 to 32.62; p = 0.785).
Figure 3.5.1 shows the risk of bias summary for studies investigating the effect of single prolonged stress on sucrose preference in animals. The risk of bias assessment was performed using the SyRCLE RoB tool.
Figure 3.5.1
Figure 3.5.2 shows the reporting completeness summary for studies investigating the effect of single prolonged stress on sucrose preference in animals. The reporting completeness assessment was performed using the ARRIVE guidelines.
Figure 3.5.2
The effect of single prolonged stress on BDNF in animals using SMD as the effect size is shown in Figure 3.5.3. The pooled estimate for SMD across all individual comparisons is displayed as a diamond shape at the bottom of the plot. Dotted lines indicate the prediction interval of the pooled estimate.
Figure 3.5.3
Single prolonged stress had a pooled effect on BDNF of SMD = -2.11, (95% CI: -4.56 to 0.34; 95% PrI: -9.3 to 5.08).
20 experimental comparisons were reported in 11 experiments from 8 publications and involving 2 different animal strain(s).
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 8 | 8.16 |
| Study x Experiment | 11 | 0 |
For each outcome, the covariates of interest for subgroup analyses and meta-regressions were:
We also conducted subgroup analyses using (1) SyRCLE Risk of Bias and (2) ARRIVE reporting completeness assessment scores as covariates to evaluate their influence on effect size estimates. These were not specified in the study protocol, but evaluation of risk of bias is required for the Summary of Evidence table, and no studies were considered entirely at low risk of bias or of high reporting completeness to allow such a sensitivity analysis
The significance (p-value) reported is that for a test of whether the moderators are significantly different one from another, rather than whether the effect is significantly different from 0.
Figure 3.5.4.1 displays the estimates for the pooled SMD’s when comparisons are stratified by sex of the animal. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by sex, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 3.5.4.1 - Effect of single prolonged stress on BDNF by Sex
The p-value for the association between the sex of animal groups used and outcome reported was 0.335.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 8 | 7.935 |
| Study x Experiment | 11 | 0 |
No studies met any RoB criteria
No studies met any RoB criteria
We provide a meta-regression where the number of ARRIVE items met is considered as a continuous variable.
Figure 3.5.4.2 - Effect of single prolonged stress on BDNF by ARRIVE reporting completeness
The p-value for the association between ARRIVE reporting completeness and outcome reported was 0.61.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0 |
| Study x Strain | 8 | 9.352 |
| Study x Strain x Experiment | 11 | 0 |
The table below shows which of the covariates, if any, explain some of the heterogeneity observed in the effect of modelling on BDNF. We present marginal R2, which measures the proportion of variance explained by including moderators in the model (the % change in the between-studies variance when the covariate is included in the model, in other words the % of the heterogeneity explained by the variable). The coefficients are derived form an RMA model fitted with an intercept (and so represent, for each category, the point estimate and 95% CIs of the effect in that category).
| Moderator | Category | \(\beta\) | 95% CI | Marginal R2 (%) |
|---|---|---|---|---|
| Overall effect | - | -2.111 | -4.561 to 0.339 | - |
| Sex | - | - | - | 0.6% |
| - | Female | -1.869 | -4.241 to 0.503 | - |
| - | Male | -2.301 | -4.652 to 0.051 | - |
| Reporting completeness | - | - | - | 3.4% |
| - | per unit increase | 0.382 | -1.357 to 2.121 | - |
We examine the robustness of the findings for BDNF by performing the following sensitivity analyses.
In the previous analyses for the effect of single prolonged stress on BDNF, we imputed a \(\rho\) value - the imputed within-study correlation between observed effect sizes - of 0.5. Here, we examine the effect of imputing \(\rho\) values of 0.2 and 0.8.
When the \(\rho\) value is assumed to be 0.2, single prolonged stress had a pooled effect on BDNF of SMD = -2.48 (95% CI: -5.13 to 0.17; 95% PrI: -10.3 to 5.33).
When the \(\rho\) value is assumed to be 0.8, single prolonged stress had a pooled effect on BDNF of SMD = -1.39 (95% CI: -3.57 to 0.79; 95% PrI: -7.83 to 5.04).
For reference the pooled effect size when rho is assumed to be 0.5 is -2.11 (95% CI: -4.56 to 0.34).
Because of the relationship between SMD effect sizes and variance inherent in their calculation, where study size is small the standard approach to seeking evidence of small-study effects (regression based tests including Egger’s regression test for multilevel meta-analysis) can lead to over-estimation of small-study effect (see for instance 10.7554/eLife.24260). To address this we used Egger’s regression test for multilevel meta-analysis, with regression of SMD effect size against 1/√N, where N is the total number of animals involved in an experiment.
Egger regression based on 20 studies of modelling of depression where sucrose preference was measured showed a coefficient for a small-study effect of 36.06 (95% CI: -41.43 to 113.55; p = 0.298).
Figure 3.6.1 shows the risk of bias summary for studies investigating the effect of model induction on biological stress response in animals. The risk of bias assessment was performed using the SyRCLE RoB tool.
Figure 3.6.1
Figure 3.6.2 shows the reporting completeness summary for studies investigating the effect of model induction on biological stress response in animals. The reporting completeness assessment was performed using the ARRIVE guidelines.
Figure 3.6.2
The effect of single prolonged stress on the biological stress response in animals using SMD as the effect size is shown in Figure 3.6.3. The pooled estimate for SMD across all individual comparisons is displayed as a diamond shape at the bottom of the plot. Dotted lines indicate the prediction interval of the pooled estimate.
Figure 3.6.3
Single prolonged stress had a pooled effect on biological stress response of SMD = -3.41, (95% CI: -5.32 to -1.51; 95% PrI: -8.13 to 1.3).
9 experimental comparisons were reported in 8 experiments from 6 publications and involving 1 different animal strain(s).
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 6 | 2.82 |
| Study x Experiment | 8 | 0 |
For each outcome, the covariates of interest for subgroup analyses and meta-regressions were:
We also conducted subgroup analyses using (1) SyRCLE Risk of Bias and (2) ARRIVE reporting completeness assessment scores as covariates to evaluate their influence on effect size estimates. These were not specified in the study protocol, but evaluation of risk of bias is required for the Summary of Evidence table, and no studies were considered entirely at low risk of bias or of high reporting completeness to allow such a sensitivity analysis.
The significance (p-value) reported is that for a test of whether the moderators are significantly different one from another, rather than whether the effect is significantly different from 0.
Figure 3.6.4.1 displays the estimates for the pooled SMD’s when comparisons are stratified by sex of the animal. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by sex, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 3.6.4.1 - Effect of single prolonged stress on biological stress response by Sex
The p-value for the association between the sex of animal groups used and outcome reported was 0.932.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 6 | 2.557 |
| Study x Experiment | 8 | 0.501 |
Figure 3.6.4.2 displays the estimates for the pooled SMD when comparisons are stratified by how many of the SyRCLE risk of bias assessment criteria (of which there are 10) that the experiment met. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by SyRCLE Risk of Bias, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 3.6.4.2 - Effect of single prolonged stress on biological stress response by SyRCLE RoB
The p-value for the association between SyRCLE Risks of Bias reporting and outcome reported was 0.414.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 6 | 2.893 |
| Study x Experiment | 8 | 0 |
Figure 3.6.4.3 displays the estimates for the pooled SMD when comparisons are stratified by whether of not any of the SyRCLE Risk of bias domains were rated as low risk of bias. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by SyRCLE Risk of Bias, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 3.6.4.3 - Effect of single prolonged stress on biological stress response by low SyRCLE RoB
The p-value for the association between low SyRCLE Risks of Bias reporting and outcome reported was 0.414.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 6 | 2.893 |
| Study x Experiment | 8 | 0 |
We provide a meta-regression where the number of ARRIVE items met is considered as a continuous variable.
Figure 3.6.4.4 - Effect of single prolonged stress on biological stress response by ARRIVE reporting completeness
The p-value for the association between ARRIVE reporting completeness and outcome reported was 0.688.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 1 | 0 |
| Study x Strain | 6 | 3.419 |
| Study x Strain x Experiment | 8 | 0 |
The table below shows which of the covariates, if any, explain some of the heterogeneity observed in the effect of model induction on biological stress response. We present marginal R2, which measures the proportion of variance explained by including moderators in the model (the % change in the between-studies variance when the covariate is included in the model, in other words the % of the heterogeneity explained by the variable). The coefficients are derived form an RMA model fitted with an intercept (and so represent, for each category, the point estimate and 95% CIs of the effect in that category).
| Moderator | Category | \(\beta\) | 95% CI | Marginal R2 (%) |
|---|---|---|---|---|
| Overall effect | - | -3.414 | -5.319 to -1.509 | - |
| Sex | - | - | - | 0.1% |
| - | Female | -3.407 | -5.907 to -0.908 | - |
| - | Male | -3.5 | -5.551 to -1.449 | - |
| Risk of Bias | - | - | - | 12.3% |
| - | 0 criteria met | -3.128 | -5.386 to -0.869 | - |
| - | 1 criteria met | -5.038 | -10.398 to 0.323 | - |
| Reporting completeness | - | - | - | 3.2% |
| - | per unit increase | 0.203 | -1.098 to 1.503 | - |
We examine the robustness of the findings for biological stress response by performing the following sensitivity analyses.
In the previous analyses for the effect of single prolonged stress on biological stress response, we imputed a \(\rho\) value - the imputed within-study correlation between observed effect sizes - of 0.5. Here, we examine the effect of imputing \(\rho\) values of 0.2 and 0.8.
When the \(\rho\) value is assumed to be 0.2, single prolonged stress had a pooled effect on biological stress response of SMD = -3.27 (95% CI: -5.44 to -1.1; 95% PrI: -8.73 to 2.19).
When the \(\rho\) value is assumed to be 0.8, single prolonged stress had a pooled effect on biological stress response of SMD = -3.78 (95% CI: -5.31 to -2.26; 95% PrI: -7.43 to -0.14).
For reference the pooled effect size when rho is assumed to be 0.5 is -3.41 (95% CI: -5.32 to -1.51).
Because of the relationship between SMD effect sizes and variance inherent in their calculation, where study size is small the standard approach to seeking evidence of small-study effects (regression based tests including Egger’s regression test for multilevel meta-analysis) can lead to over-estimation of small-study effect (see for instance 10.7554/eLife.24260). To address this we used Egger’s regression test for multilevel meta-analysis, with regression of SMD effect size against 1/√N, where N is the total number of animals involved in an experiment.
Egger regression based on 9 studies of modelling of depression where sucrose preference was measured showed a coefficient for a small-study effect of 22.53 (95% CI: -84.41 to 129.47; p = 0.590).
Multilevel analysis is only performed if there are 5 levels or more for at least one of Strain, Study and Experiment, and that is not the case here. 10 experimental comparisons were reported in 1 experiments from 1 publications and involving 1 different animal strain(s). We provide a conventional random effects model to illustrate the data. No subgroup analysis is performed. We also now include these freezing outcomes under the ‘other neurobiological’ heading below.
Figure 3.8.1 shows the risk of bias summary for studies investigating the effect of single prolonged stress on other neurobiological outcomes in animals. The risk of bias assessment was performed using the SyRCLE RoB tool.
Figure 3.8.1
Figure 3.8.2 shows the reporting completeness summary for studies investigating the effect of single prolonged stress on other neurobiological outcomes in animals. The reporting completeness assessment was performed using the ARRIVE guidelines.
Figure 3.8.2
The effect of single prolonged stress on other neurobiological outcomes in animals using SMD as the effect size is shown in Figure 3.1.3. The pooled estimate for SMD across all individual comparisons is displayed as a diamond shape at the bottom of the plot. Dotted lines indicate the prediction interval of the pooled estimate.
Figure 3.8.3
Single prolonged stress had a pooled effect on other neurobiological outcomes of SMD = -1.68, (95% CI: -3.82 to 0.46; 95% PrI: -7.92 to 4.56).
50 experimental comparisons were reported in 9 experiments from 8 publications and involving 2 different animal strain(s).
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 8 | 6.14 |
| Study x Experiment | 9 | 0 |
For each outcome, the covariates of interest for subgroup analyses and meta-regressions were:
We also conducted subgroup analyses using (1) SyRCLE Risk of Bias and (2) ARRIVE reporting completeness assessment scores as covariates to evaluate their influence on effect size estimates. These were not specified in the study protocol, but evaluation of risk of bias is required for the Summary of Evidence table, and no studies were considered entirely at low risk of bias or of high reporting completeness to allow such a sensitivity analysis.
The significance (p-value) reported is that for a test of whether the moderators are significantly different one from another, rather than whether the effect is significantly different from 0.
Figure 3.8.4.1 displays the estimates for the pooled SMD when comparisons are stratified by sex of the animal. Whiskers indicate the 95% confidence interval of each estimate. The overall pooled SMD, not stratified by sex, is displayed as a diamond shape at the bottom of the plot, with the 95% prediction interval shown as a red bar.
Figure 3.8.4.1 - Effect of single prolonged stress on other neurobiological outcomes by Sex
The p-value for the association between the sex of animal groups used and outcome reported was 0.335.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Study | 8 | 6.539 |
| Study x Experiment | 9 | 0 |
No studies met any RoB criteria.
No studies met any RoB criteria.
We provide a meta-regression where the number of ARRIVE items met is considered as a continuous variable.
Figure 3.8.4.2 - Effect of single prolonged stress on other neurobiological outcomes by ARRIVE reporting completeness
The p-value for the association between ARRIVE reporting completeness and outcome reported was 0.233.
| Level | Number of categories for that level included in this analysis | Attributable variance |
|---|---|---|
| Strain | 2 | 0 |
| Study x Strain | 8 | 6.001 |
| Study x Strain x Experiment | 9 | 0 |
The table below shows which of the covariates, if any, explain some of the heterogeneity observed in the effect of modelling on other neurobiological outcomes. We present marginal R2, which measures the proportion of variance explained by including moderators in the model (the % change in the between-studies variance when the covariate is included in the model, in other words the % of the heterogeneity explained by the variable). The coefficients are derived form an RMA model fitted with an intercept (and so represent, for each category, the point estimate and 95% CIs of the effect in that category).
| Moderator | Category | \(\beta\) | 95% CI | Marginal R2 (%) |
|---|---|---|---|---|
| Overall effect | - | -1.681 | -3.821 to 0.46 | - |
| Sex | - | - | - | 0.4% |
| - | Female | -1.864 | -4.256 to 0.527 | - |
| - | Male | -1.554 | -3.878 to 0.771 | - |
| Reporting completeness | - | - | - | 14.6% |
| - | per unit increase | 0.652 | -0.551 to 1.856 | - |
We examine the robustness of the findings for other neurobiological outcomes by performing the following sensitivity analyses
In the previous analyses for the effect of single prolonged stress on other neurobiological outcomes, we imputed a \(\rho\) value - the imputed within-study correlation between observed effect sizes - of 0.5. Here, we examine the effect of imputing \(\rho\) values of 0.2 and 0.8.
When the \(\rho\) value is assumed to be 0.2, single prolonged stress had a pooled effect on other neurobiological outcomes of SMD = -2.38 (95% CI: -4.29 to -0.47; 95% PrI: -7.94 to 3.18).
When the \(\rho\) value is assumed to be 0.8, single prolonged stress had a pooled effect on other neurobiological outomes of SMD = -0.53 (95% CI: -3.2 to 2.13; 95% PrI: -8.38 to 7.32).
For reference the pooled effect size when rho is assumed to be 0.5 is -1.68 (95% CI: -3.82 to 0.46).
Because of the relationship between SMD effect sizes and variance inherent in their calculation, where study size is small the standard approach to seeking evidence of small-study effects (regression based tests including Egger’s regression test for multilevel meta-analysis) can lead to over-estimation of small-study effect (see for instance 10.7554/eLife.24260). To address this we used Egger’s regression test for multilevel meta-analysis, with regression of SMD effect size against 1/√N, where N is the total number of animals involved in an experiment.
Egger regression based on 50 studies of modelling of depression where other neurobiological outcomes was measured showed a coefficient for a small-study effect of 26.02 (95% CI: -39.67 to 91.72; p = 0.370).
We selected cohorts where at least one outcome was presented for at least two outcome types. Where there were two or more of the same outcome type within a cohort, we calculated a standardised mean difference effect size for that outcome in that cohort, along with its standard error. Where there was a single effect size within a cohort, we took the standard error of that effect size.
Then, for each pair of outcome measures we plotted the effect sizes for each cohort, and fitted a regression line weighted on the standard error in the outcome measure represented on the x-axis. Outcome measure pairs are coded according to whether they come from model induction studies (red, expectation of worsening anhedonia) or from intervention studies (green, expectation of improvement in anhedonia). The number of experimental comparisons observed from each cohort is reflected in the size of the symbol, and shown in the figure legend.
We show pairwise relationships where there were at least 3 experimental cohorts for either modelling experiments, intervention experiments, or both, and provide the relevant regression coefficient(s). We also provide an overall representation of all paired neurobiological effect sizes with the corresponding behavioural outcome in that cohort.
4.1 Relationship between change in locomotor activity and change in ‘other’ neurobiological outcomes
4.2 Relationship between change in freezing behaviour and change in ‘other’ behavioural outcomes
4.3 Relationship between change in BDNF and change in ‘other’ behavioural outcomes
4.4 Relationship between change in stress response and change in ‘other’ behavioural outcomes
4.5 Relationship between change in ‘other’ neurobiological outcomes and change in ‘other’ behavioural outcomes
4.6 Relationship between change in stress response and change in BDNF
4.7 Relationship between change in ‘other’ neurobiological outcomes and change in BDNF
4.8 Relationship between change in ‘other’ neurobiological outcomes and change in stress response
4.9 Relationship between change in all neurobiological outcomes and all behavioural outcomes
We know of 1823 animals in Control cohorts and 1823 animals in Intervention cohorts, of which 0% and 0% were reported to have ‘dropped out’ between allocation to group and outcome measurement. This analysis assumes full reporting of animals excluded from analyses, and it may be that group sizes were specified ‘after the event’, or that there was unreported replacement of animals excluded during the experiment, so these data should be interpreted with caution.
| Outcome | Summary of the association | Within-study biases | Across-studies biases | Indirectness | Other biases |
|---|---|---|---|---|---|
| Locomotor activity | 13 experimental comparisons were reported in 4 experiments from 4 publications involving 2 different animal strains and reporting data from 260 animals. Univariate metaregression performed; SMD = 1.30 (95% CI to 0.74 to 1.87, 95% PrI -0.47 to 3.08) (Section 2.1). | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 13.5 (of 22). | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. | Moderate risk of indirectness. For explanation, see [1] below. | No other risks identified. |
| Fear Memory | 7 experimental comparisons were reported in 4 experiments from 4 publications involving 1 animal strain and reporting data from 110 animals. Univariate metaregression performed; SMD = 2.55 (95% CI 1.29 to 3.82; 95% PrI -0.54 to 5.65) (Section 2.2). | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 13 (of 22). | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. | Moderate risk of indirectness. For explanation, see [1] below. | No other risks identified. |
| Freezing | 4 experimental comparisons were reported in 3 experiments from 3 publications involving 2 different animal strains and reporting data from 64 animals. Univariate metaregression performed; SMD = 1.06 (95% CI 0.20 to 1.93; 95% PrI 0.18 to 1.94) (Section 2.3). | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 15 (of 22).C5 | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. | Moderate risk of indirectness. For explanation, see [1] below. | No other risks identified. |
| All behaviours | 74 experimental comparisons were reported in 19 experiments from 14 publications involving 2 different animal strains and reporting data from 1192 animals. SMD = 1.33 (95% CI 0.89 to 1.78; 95% PrI -0.24 to 2.90) (Section 2.4). The effect was lower with increasing total exercise (n sessions x duration x speed).B4 | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 14.5 (of 22). | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. There was no evidence of small study effects. | Moderate risk of indirectness. For explanation, see [1] below. | No other risks identified. |
| BDNF | 24 experimental comparisons were reported in 11 experiments from 8 publications involving 2 different animal strains and reporting data from 400 animals. SMD = 1.79 (95% CI 0.56 to 3.01; 95% PrI -1.63 to ) (Section 2.5). The effect was lower with increasing exercise intensity and with total exercise (n sessions x duration x speed). | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 14.5 (of 22). | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. There was no evidence of small study effects. | Moderate risk of indirectness. For explanation, see [2] below. | No other risks identified. |
| Stress response | 10 experimental comparisons were reported in 8 experiments from 6 publications involving 1 animal strains and reporting data from 170 animals. SMD = 2.03 (95% CI -1.78 to 5.84; 95% PrI -7.88 to 11.93) (Section 2.6). The effect was lower with increasing exercise intensity and with increasing total exercise (n sessions x duration x speed). | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 14.5 (of 22). | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. There was no evidence of small study effects. | Moderate risk of indirectness. For explanation, see [2] below. | No other risks identified. |
| Neurotransmitter levels | 10 experimental comparisons were reported in 1 experiments from 1 publications involving 1 animal strain and reporting data from 200 animals. SMD = 0.07 (95% CI -0.32 to 0.46; 95% PrI -0.70 to 0.83 ) (Section 2.7). | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 15 (of 22). | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. | Moderate risk of indirectness. For explanation, see [2] below. | No other risks identified. |
| All neurobiological except BDNF and stress response | 62 experimental comparisons were reported in 11 experiments from 9 publications involving 2 different animal strains and reporting data from 998 animals. SMD = 1.10 (95% CI -0.18 to 2.38; 95% PrI -2.79 to 4.99 ). The effect was lower with increasing exercise intensity (Section 2.8). | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 15 (of 22). | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. There was no evidence of small study effects. | Moderate risk of indirectness. For explanation, see [3] below. | No other risks identified. |
To provide context for the effects of exercise in these models, we also present a summary of the effects of model induction.
| Outcome | Summary of the association | Within-study biases | Across-studies biases | Indirectness | Other biases |
|---|---|---|---|---|---|
| Locomotor activity | 13 experimental comparisons were reported in 4 experiments from 4 publications involving 2 different animal strains and reporting data from 260 animals. Univariate metaregression performed; SMD = -1.30 (95% CI -1.83 to 0.78, 95% PrI -2.92 to 0.32) (Section 3.1). | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 13.5 (of 22). | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. | Moderate risk of indirectness. For explanation, see [3] below. | No other risks identified. |
| Fear Memory | 7 experimental comparisons were reported in 4 experiments from 4 publications involving 1 animal strain and reporting data from 110 animals. Univariate metaregression performed; SMD = -2.55 (95% CI -4.40 to -0.70; 95% PrI -7.46 to 2.36) (Section 3.2). | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 13 (of 22). | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. | Moderate risk of indirectness. For explanation, see [3] below. | No other risks identified. |
| Freezing | 4 experimental comparisons were reported in 3 experiments from 3 publications involving 2 different animal strains and reporting data from 52 animals. Univariate metaregression performed; SMD = -2.44 (95% CI -5.96 to 1.08; 95% PrI -8.79 to 3.91) (Section 3.3). | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 15 (of 22). | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. | Moderate risk of indirectness. For explanation, see [3] below. | No other risks identified. |
| All behaviours | 64 experimental comparisons were reported in 19 experiments from 14 publications involving 2 different animal strains and reporting data from 1064 animals. SMD = -1.64 (95% CI -2.50 to -0.78; 95% PrI -4.79 to 1.50) (Section 3.4). | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 14.5 (of 22). | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. There was no evidence of small study effects. | Moderate risk of indirectness. For explanation, see [3] below. | No other risks identified. |
| BDNF | 20 experimental comparisons were reported in 11 experiments from 8 publications involving 2 different animal strains and reporting data from 344 animals. SMD = -2.11 (95% CI -4.56 to 0.34; 95% PrI –9.30 to 5.08) (Section 3.5). | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 14.5 (of 22). | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. There was no evidence of small study effects. | Moderate risk of indirectness. For explanation, see [3] below. | No other risks identified. |
| Stress response | 9 experimental comparisons were reported in 8 experiments from 6 publications involving 1 animal strains and reporting data from 156 animals. SMD = -3.41 (95% CI -5.32 to -1.51; 95% PrI -1.02 to 1.18) (Section 3.6). | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 14.5 (of 22). | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. There was no evidence of small study effects. | Moderate risk of indirectness. For explanation, see [3] below. | No other risks identified. |
| Neurotransmitter levels | 10 experimental comparisons were reported in 1 experiments from 1 publications involving 1 animal strain and reporting data from 200 animals. SMD = 0.08 (95% CI -0.37 to 0.54; 95% PrI –0.70 to 0.83 ) (Section 3.7). | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 15 (of 22). | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. | Moderate risk of indirectness. For explanation, see [3] below. | No other risks identified. |
| All neurobiological except BDNF and stress response | 50 experimental comparisons were reported in 9 experiments from 8 publications involving 2 different animal strains and reporting data from 998 animals. SMD = -1.68 (95% CI -3.82 to 0.46; 95% PrI -7.92 to 4.56) (Section 3.8). | Moderate risk of bias likely to exaggerate the effects of exercise. All studies had unclear risk of bias for most of the SyRCLE items. Reporting was mostly incomplete; the median number of ARRIVE items reported was 15 (of 22). | Moderate risk of bias likely to exaggerate the effects of exercise. No studies preregistered their analyses. There was no evidence of small study effects. | Moderate risk of indirectness. For explanation, see [3] below. | No other risks identified. |
Rationale for conclusions for indirectness:
[1] Effect of exercise on behavioural outcomes in PTSD: Moderate risk of indirectness Locomotor activity is increased following treatment, but in humans hyperlocomotion may be observed as a feature of PTSD, and is thought to reflect agitation. Other behavioural outcomes reported (fear memory, freezing, anxiety and cognitive impairment) have good Ethological validity, with clinically relevant human correlates.
[2] Effect of exercise on neurobiological outcomes in PTSD: Moderate risk of indirectness BDNF levels are increased following treatment, and while PTSD is associated with increased BDNF levels in humans, there is also a reported association between high levels of BDNF and greater response to treatment. Similarly, changes in stress physiology are seen here and are associated with improved outcomes in human PTSD. We saw no significant change in neurotransmitter levels. Only one study reported serotonin levels, showing reduced hippocampal and prefrontal cortex serotonin (5-HT) following SPS, with reversal of hippocampal but not prefrontal cortex following treatment. Serum serotonin levels are reportedly reduced in people with PTSD.
[3] Validity of the single prolonged stress model: Mild to moderate risk of indirectness The single prolonged stress model has reasonable Homological, Triggering, Induction, and Remission validity. No studies included an early ‘priming’ exposure which might give Ontopathogenic validity, and there was no significant effect of exercise on Freezing behaviour (although the number of included studies was low, and the point estimate favoured an effect). The elevated levels of BDNF seen in human PTSD were not seen in animal models (there was a non significant reduction in BDNF levels), when tested 4 weeks after induction, giving reduced Biomarker validity.Markers of physiological stress were increased by SPS, consistent with findings in humans; and there was no effect on neurotransmiter levels or on other neurobiological markers such as the abundance of Bax or Bcl-2.
The framework for the evaluation of indirectness is based on eight dimensions, based on the work of Belzung and Lemoine, and comprising (i) Homological validity - what is the extent of homology between the model organism and humans relevant to the condition studied; (ii) Ontopathogenic validity - Does the model include prenatal or early life exposures inducing transition from initial organism to vulnerable organism; (iii) Triggering validity - are any triggering factors used in the modelling – or their homologues -known to induce psychosis or relapse in humans?; (iv) Mechanistic validity - whether the neurobiological or cognitive mechanisms which operate in human disease can be observed in the animal model; (v) Induction validity - Does the induction of the disease model induce changes in biomarkers (see below) which are known to be altered in human disease?; (vi) Remission validity - What is the effect of other drugs known to be effective in humans in the particular animal model / outcome measure pair? ; (vii) Biomarker validity - are changes in disease markers (eg neurotransmitter levels, structural brain imaging) seen in human disease also seen in this animal model?; and (viii) Ethological validity - what is the ‘behavioural distance’ between the model phenotype in animals and the symptoms and signs of human disease at which treatment is targeted?
| Dimension | Characteristic | Homological validity | Ontopathogenic validity | Triggering validity | Mechanistic validity | Induction validity | Remission validity | Biomarker validity | Ethological validity |
|---|---|---|---|---|---|---|---|---|---|
| Species and strain | Rat | All studies were conducted in rats, which do manifest behavioural responses to stress | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. |
| Model Induction | Single prolonged stress (SPS) | SPS induces changes in rat behaviours with some homology to changes seen in PTSD including negative alterations in cognition and alterations in arousal and reactivity. | None of the experiments included early ‘priming’ exposures | SPS induces changes in several behaviours including locomotor activity, fear memory, and other behaviours (Sections 3.1, 3.2, 3.4). There was no significant effect on freezing (Section 3.3) | SPS induces changes in stress response, but no significant changes in BDNF, neurotransmitter levels, or other neurobehavioural outcomes (Sections 3.5:3.8) | Intra-operative awareness (10.1016/j.genhosppsych.2021.01.010), Near drowning (PMID: 19289877) and physical restraint in psychiatric inpatients (10.3389/fpsyt.2019.00491) are each associated with PTSD | multimodal exercise is effective in human PTSD (10.1177/02692155231225466), and exercise is effective in rat SPS models (Section 2) | BDNF levels are reportedly increased in human PTSD (10.1016/j.pnpbp.2024.110954), and not significantly changed in animal models (Section 3.5) considered here (measured at median 4 weeks after SPS) . Other reports suggest increased BDNF following stressful exposures in animals. | n.a. |
| Outcome Measure | Locomotor activity | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | Hyperlocomotion in humans is taken as a marker of agitation, but in animal models of PTSD reduced activity in the open field test is held to represent a disease phenotype |
| ~ | Fear Memory | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | Avoidance of trauma-related stimuli after the trauma is a DSM-5-TR diagnostic requirement for human PTSD |
| ~ | Freezing | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | Tonic immobility is a feature of human PTSD |
| ~ | Other behavioural: Includes tests of anxiety (elevated plus maze), cognition (novel object recognition task, radial arm water maze), | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | Anxiety is a feature of PTSD; cognitive impairment is a feature of PTSD (/doi.org/10.1037/a0038039) |
| ~ | BDNF | n.a. | n.a. | n.a. | n.a. | n.a. | high levels of BDNF are associated with treatment response in humans (10.1016/j.jpsychires.2024.02.045), and the BDNF response is associated with behavioural response to exercise here (Section 4.3). | n.a. | n.a. |
| ~ | Stress response | n.a. | n.a. | n.a. | n.a. | n.a. | A more dynamic response in the dexamethasone suppression test was associated with improved outcomes in a trial of trauma focussed psychotherapy for PTSD (10.1016/j.jad.2015.05.058) | n.a. | NA |
| ~ | Neurotransmitter levels | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | serum serotonin levels are reduced in people with PTSD (10.3390/ijerph192416517) and animal hippocampal 5-HT levels are reduced following SPS and increased following exercise |
| ~ | Other neurobiological: markers of apoptosis and neurodegeneration | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | n.a. | There are some human data suggesting hiipocampal volume loss, but no data on markers of apoptosis |
The description of the criteria is available at https://doi.org/10.17605/OSF.IO/TDMAU
We were only able to evaluate this where there were data for pairs of outcome measures from at least three different experimental cohorts. Mapping all behavioural outcomes (which might be considered ‘apical’ or more clinically relevant) against all neurobiological outcomes (which might be considered intermediate or mechanistic outcomes), there was reasonable correlation (r2 = 0.37) in effects of model induction, but not for the effects of exercise (r2 = 0.07).
There were insufficient data to compare different behavioural outcome measures except for freezing behaviour compared with other behaviours for exercise treatment, where there was no relationship (r2 = 0.02).
Within Neurobiological responses, there was good agreement between effects on BDNF and effects on the stress response for both model induction (r2 = 0.43) and exercise treatment (r2 = 0.81); between effects on BDNF and effects on other neurobiological outcomes for both model induction (r2 = 0.49) and exercise treatment (r2 = 0.59); and between effects on the stress response and effects on other neurobiological outcomes for both model induction (r2 = 0.27) and exercise treatment (r2 = 0.97).
Finally, we mapped individual neurobiological (‘intermediate’) outcomes against all behavioural outcomes (not shown). There was some agreement between effects on BDNF and effects on all behavioural outcomes for both model induction (r2 = 0.21) and exercise treatment (r2 = 0.25) and between effects on the stress response and effects on all behavioural outcomes for both model induction (r2 = 0.29) and exercise treatment (r2 = 0.14).
We used R version 4.3.1 (R Core Team 2023) and the following R packages: devtools v. 2.4.5 (Wickham et al. 2022), dosresmeta v. 2.0.1 (Crippa and Orsini 2016), ggpubr v. 0.6.0 (Kassambara 2023), gtools v. 3.9.5 (Warnes et al. 2023), Hmisc v. 5.1.1 (Harrell Jr 2023a), kableExtra v. 1.4.0.3 (Zhu 2024), knitr v. 1.45 (Xie 2014, 2015, 2023), Matrix v. 1.6.5 (Bates, Maechler, and Jagan 2024), meta v. 7.0.0 (Balduzzi, Rücker, and Schwarzer 2019), metadat v. 1.2.0 (White et al. 2022), metafor v. 4.4.0 (Viechtbauer 2010), mvmeta v. 1.0.3 (Gasparrini, Armstrong, and Kenward 2012), numDeriv v. 2016.8.1.1 (Gilbert and Varadhan 2019), orchaRd v. 2.0 (Nakagawa et al. 2023), patchwork v. 1.2.0 (Pedersen 2024), PRISMA2020 v. 1.1.1 (Haddaway et al. 2022), rje v. 1.12.1 (Evans 2022), rms v. 6.7.1 (Harrell Jr 2023b), robvis v. 0.3.0.900 (McGuinness and Higgins 2020), scales v. 1.3.0 (Wickham, Pedersen, and Seidel 2023), tidyverse v. 2.0.0 (Wickham et al. 2019), usethis v. 2.2.3 (Wickham et al. 2024), xtable v. 1.8.4 (Dahl et al. 2019).